[llvm] [AMDGPU] Fix implicit $vcc operands after parsing MIR (PR #87781)

[Jay Foad via llvm-commits]

================
@@ -135,42 +135,49 @@ name:            vopd_cndmask
 tracksRegLiveness: true
 body:             |
   bb.0:
+    liveins: $vcc_lo
     ; SCHED-LABEL: name: vopd_cndmask
-    ; SCHED: $vgpr2 = IMPLICIT_DEF
+    ; SCHED: liveins: $vcc_lo
+    ; SCHED-NEXT: {{  $}}
+    ; SCHED-NEXT: $vgpr2 = IMPLICIT_DEF
     ; SCHED-NEXT: $vgpr0 = IMPLICIT_DEF
     ; SCHED-NEXT: $vgpr1 = IMPLICIT_DEF
     ; SCHED-NEXT: $vgpr3 = IMPLICIT_DEF
     ; SCHED-NEXT: $sgpr20 = IMPLICIT_DEF
     ; SCHED-NEXT: $vgpr4 = V_FMAMK_F32 $sgpr20, 12345, $vgpr3, implicit $mode, implicit $exec
     ; SCHED-NEXT: $vgpr2 = V_FMAC_F32_e32 $sgpr20, killed $vgpr1, killed $vgpr2, implicit $mode, implicit $exec
-    ; SCHED-NEXT: $vgpr5 = V_CNDMASK_B32_e32 $vgpr0, $vgpr3, implicit $mode, implicit $exec, implicit $vcc
-    ; SCHED-NEXT: $vgpr7 = V_CNDMASK_B32_e32 killed $vgpr0, $vgpr3, implicit $mode, implicit $exec, implicit $vcc
+    ; SCHED-NEXT: $vgpr5 = V_CNDMASK_B32_e32 $vgpr0, $vgpr3, implicit $mode, implicit $exec, implicit $vcc_lo
+    ; SCHED-NEXT: $vgpr7 = V_CNDMASK_B32_e32 killed $vgpr0, $vgpr3, implicit $mode, implicit $exec, implicit $vcc_lo
     ; SCHED-NEXT: $vgpr6 = V_ADD_F32_e32 $sgpr20, $vgpr3, implicit $mode, implicit $exec
-    ; SCHED-NEXT: $vgpr9 = V_CNDMASK_B32_e32 killed $sgpr20, killed $vgpr3, implicit $mode, implicit $exec, implicit $vcc
+    ; SCHED-NEXT: $vgpr9 = V_CNDMASK_B32_e32 killed $sgpr20, killed $vgpr3, implicit $mode, implicit $exec, implicit killed $vcc_lo
     ;
     ; PAIR-GFX11-LABEL: name: vopd_cndmask
-    ; PAIR-GFX11: $vgpr2 = IMPLICIT_DEF
+    ; PAIR-GFX11: liveins: $vcc_lo
+    ; PAIR-GFX11-NEXT: {{  $}}
+    ; PAIR-GFX11-NEXT: $vgpr2 = IMPLICIT_DEF
     ; PAIR-GFX11-NEXT: $vgpr0 = IMPLICIT_DEF
     ; PAIR-GFX11-NEXT: $vgpr1 = IMPLICIT_DEF
     ; PAIR-GFX11-NEXT: $vgpr3 = IMPLICIT_DEF
     ; PAIR-GFX11-NEXT: $sgpr20 = IMPLICIT_DEF
     ; PAIR-GFX11-NEXT: $vgpr4 = V_FMAMK_F32 $sgpr20, 12345, $vgpr3, implicit $mode, implicit $exec
-    ; PAIR-GFX11-NEXT: $vgpr2, $vgpr5 = V_DUAL_FMAC_F32_e32_X_CNDMASK_B32_e32_gfx11 $sgpr20, killed $vgpr1, killed $vgpr2, $vgpr0, $vgpr3, implicit $mode, implicit $exec, implicit $vcc, implicit $mode, implicit $exec, implicit $mode, implicit $exec, implicit $vcc
-    ; PAIR-GFX11-NEXT: $vgpr7 = V_CNDMASK_B32_e32 killed $vgpr0, $vgpr3, implicit $mode, implicit $exec, implicit $vcc
+    ; PAIR-GFX11-NEXT: $vgpr2, $vgpr5 = V_DUAL_FMAC_F32_e32_X_CNDMASK_B32_e32_gfx11 $sgpr20, killed $vgpr1, killed $vgpr2, $vgpr0, $vgpr3, implicit $mode, implicit $exec, implicit $vcc, implicit $mode, implicit $exec, implicit $mode, implicit $exec, implicit $vcc_lo
----------------
jayfoad wrote:

It seems wrong that these V_DUAL_* instructions now have both "implicit $vcc" and "implicit $vcc_lo" but I don't think it's the fault of this patch. I will investigate.

https://github.com/llvm/llvm-project/pull/87781